Comparison of Machine Learning Methods for Prediction of Epilepsy by Neurophysiological EEG Signals

Abstract

Investigation of brain disorders especially epilepsy and impaired cognitive functions are the most common clinical application of neurophysiologic signals. EEG signals reflect the activity of brain and are capable of assessing the brain condition during abnormalities. In this study we have investigated the potential of two different algorithms (back propagation and radial basis function) of neural network technique for classification of patients suffering from epilepsy through EEG. Classification is based on quantitative parameters obtained from neurophysiologic signals used to train the networks and the performance of the networks is analyzed to confirm the efficacy of the network. Accuracy obtained with multi-layer perceptron NN is 99.6% and with radial basis function is 96.8%. The sensitivity obtained for pre-ictal, ictal and normal conditions are 93.9%, 100% and 97%, respectively in case of back propagation neural network algorithm. The comparative analysis is based on variation in network topology and in feature vector used for training the networks. Results from this study indicate that a classification system based on ANN may help in automation of analysis of neurophysiologic signals and the number and type of parameters used as feature set decide the type of network to be used for the better efficiency of the system.